Putting the Human Into Artificial Intelligence

Why artificial intelligence for B2B sales requires human-based fuel

By Greg Dvorken

Working in the software industry, the terms AI (artificial intelligence) and ML (machine learning) are thrown around quite a bit. It’s a set of buzz words that have captured the imagination. The silver bullets. The panaceas. The holy grails.

From what I know about effective AI and ML, it’s based on healthy inputs. Garbage in — garbage out. You must teach AI and ML based on strong data points that allow it to learn properly as more data is applied over time. These intelligent systems are designed to augment or enhance the human, serving as a tool to be wielded through human interaction.

If the wrong data and learning process is applied out of the gate, your trajectory will be off and you and your customers will end up in an unintended place.

And when you hear the terms AI and ML applied to my world, which is sales intelligence software for B2B, these basic rules absolutely apply.

What happens when you apply the wrong data to your AI and ML

The sales and marketing disciplines have been collecting data for years. Metrics such as likes, clicks, open rates, page views and impressions have been the norm for a while. They have spawned industries in analytics to better understand the value of this data, including traffic flow, customer journeys and advertising targets and rates. If you’re looking at generating brand awareness and qualified marketing leads for a consumer product or service with a short/transactional sales cycle, these metrics can tell you that kind of a story.

In the B2B world, many try to apply these same data points to create a shortcut to customer intelligence, even down-funnel after a prospect becomes a sales qualified lead. It seems easy. However, this process could involve hundreds of interactions across multiple buyer personas over a long sales cycle. That’s difficult. So then there’s going back to measuring clicks, which is super easy. But what does a click really tell you about your B2B customer’s buying behavior?

Unfortunately, in B2B enterprise sales, not enough. This is an inherently flawed starting point for B2B sales intelligence. This is because a click or a like doesn’t get to the “why” behind the B2B customer action.

Trying to definitively say a business outcome (a positive response, a meeting, movement to the next sales milestone) has occurred based on a click is like trying to guess how a person is feeling by simply looking at their face. That would be a 100% guess based on an incorrect assumption that a view translates directly into buying intent. It’s just simply not that simple.

This approach can quickly create a lot of junk and artificial bias that doesn’t get to the heart of why a customer interaction worked or didn’t work. It sets your AI and ML on that bad trajectory mentioned at the beginning of the article.

Hand-feeding your AI and ML the right stuff

“Human-in-the-loop machine learning” is the practice of uniting human and machine intelligence to create effective machine learning algorithms. It’s a blend of supervised machine learning and active learning. Humans are involved in both the training and testing stages of building an algorithm, which creates a continuous feedback loop that allows the algorithm to produce better results each time.

The next action of whether the click resulted in a helpful result (e.g., down-funnel movement) doesn’t lie within counting the clicks themselves. It lies beneath the purpose or result before, during or after those clicks. It lies in the “why.”

Business-to-business sales is a different animal. You have a pretty well-defined group of people making decisions based on a number of different factors. These include price, technical requirements, legal and regulatory requirements, business objectives, end-customer requirements, etc. The customer/buyer personas that are involved in evaluating these factors include procurement, legal, compliance, business development, privacy and technology decision-makers.

The interesting thing about B2B sales is that it’s very much a human process. While your customers may do more of their research online, and in many cases qualify you before you qualify them, it’s still a human salesperson who takes the reins after that qualification occurs and takes the customer on the rest of the journey — all the way to a win or loss.

Collecting a human-powered data point comes through a process of active learning. This involves understanding if the deal progressed (or not) and why (context), capturing that response and then baking it into your intelligence gathering and distribution process. If that process involves AI or ML to better understand that customer and what to do next, it’s imperative you capture both the what (deal moves forward or not) and the why (context).

Another side benefit of capturing these human-powered data points is that the data now stays within the collective selling community of your organization. If this individual salesperson decides to move on for whatever reason, their insights remain behind.

Now you’re ready to get predictive

Once we agree on the right fuel, then it’s about finding a format to collect and analyze this human-powered data point, as well as a system capable of sharing those insights with the people who can benefit most from these insights at that particular moment.

Unfortunately the CRM industry hasn’t quite nailed this yet. According to Forrester at its annual B2B Summit North America in May 2021, the CRM industry, on average, has about a 13% utilization rate globally. This means that almost 90% of your salesforce isn’t using the software that was designed to make them smarter as a collective unit.

Capturing data must be simple and fast. It must not detract sellers from their flow or feel administrative, or it will be summarily rejected.

Once the direction, fuel and format are agreed upon, human-powered active learning can begin.

Once this data begins to flow, organizations must be able to quickly identify where uncertainties lie in their dataset and filter out the risk. Their human experts (in this case salespeople) must be able to easily confirm assumptions while continuing to focus on their sales opportunities, and those confirmations must then easily transfer to supporting teams from marketing, communications and operations — helping to close that loop.

If you can quickly and easily deliver real-time data capture with a human confirmation, you can now deliver on the promise of true sales intelligence. This includes making real-time adjustments to existing sales campaigns and ultimately becoming predictive in the way you serve up the right information at the right time.

This is when human-powered active learning blends into unsupervised machine learning. Human direction drives the direction of automated information curation, which then speeds up your sellers’ response times and allows individuals to start to see around corners, and scale that capability as a newly formed, collectively intelligent sales organization.

This human-powered collective intelligence then starts to create something closer to that elusive holy grail which combines focus, speed, accuracy, scale, community and a closed-loop feedback system that continuously gets smarter. That’s when a paradigm shift can occur, enabling B2B selling to truly become a team sport.

Images by Jeremy Bishop on Unsplash and the Stanford Institute for Human-Centered Artificial Intelligence

Greg is the founder and CEO of Return Solutions, Inc., a B2B sales intelligence software company based in Baltimore, Maryland.